fix: add flag for overriding classic Python search values #4195
Conversation
rwgk
changed the title
|
Actually I think it's our stuff around FindPython that's breaking, not this one. Will make a new PR just in case I'm wrong and this history is useful. |
|
No, I take that back, this is running the old one - where are these values being set? Not by FindPythonInterp as far as I can tell... |
tools/FindPythonLibsNew.cmake
Outdated
| @@ -95,9 +95,24 @@ if(NOT PythonLibsNew_FIND_VERSION) | |||
| set(PythonLibsNew_FIND_VERSION "3.6") | |||
| endif() | |||
|
|
|||
| # Save variables that get set by PythonInterp | |||
There was a problem hiding this comment.
From my viewpoint as someone not knowing much about cmake, copying the title of this PR here, ideally even with a link to this PR, would seem super helpful. Otherwise I'd give too much credit to this code looking fresh. Maybe:
# For PyPy, broken FindPythonInterp values need to be overridden (see PR #4195).
# This macro saves variables that get set by PythonInterp.
There was a problem hiding this comment.
This isn't working, so I'm going to need to address this differently anyway. I'll probably do one of two things. If I can stop these from being set at all, then the current form is fine. If not, I can introduce a toggle and then you could override all or none. Something like that. :/
There was a problem hiding this comment.
(So far sill have no idea who is (incorrectly) setting these values, it's not FindPythonInterp and I'm pretty sure we don't run FindPythonLibs)
henryiii
force-pushed
the
henryiii/fix/pypy
branch
2 times, most recently
from
3e31783
to
c2eb69f
Compare
henryiii
changed the title
Signed-off-by: Henry Schreiner <henryschreineriii@gmail.com>
Signed-off-by: Henry Schreiner <henryschreineriii@gmail.com>
henryiii
force-pushed
the
henryiii/fix/pypy
branch
from
c2eb69f
to
e0b0766
Compare
|
@rwgk or @Skylion007 What do you think? This reverts the 2.10 default change and instead makes it opt-in for now. The problem is that currently tools like scikit-build reuse the same cache dir for multiple builds, which causes the "promote to cache" that's done here to override new values with old ones on new runs. Scikit-build really should make per-interpreter cached directories; that would help, but for now, this seems like the simplest fix. However, I'm torn - I'm also fine to say it's scikit-build's problem, and it needs include the interpreter string in the build path that gets cahced. Scikit-build-core will probably be designed this way from the start (currently, it just doesn't support caching at all - caching build directories is hard 😆). But that will make fixing scikit_build_example ugly for a bit until a proper fix lands in scikit-build. But this might motivate me. ;) What do you think? Option a) (this PR) or option b) (declare it not broken and working as expected). |
* fix: PyPy needs to overrite broken FindPythonInterp values Signed-off-by: Henry Schreiner <henryschreineriii@gmail.com> * fix: add flag to opt-in to new (cross-compile) behavior Signed-off-by: Henry Schreiner <henryschreineriii@gmail.com> * Apply suggestions from code review Signed-off-by: Henry Schreiner <henryschreineriii@gmail.com>
Description
Should fix pybind/scikit_build_example#45.
Suggested changelog entry:
* CMake: revert overwrite behavior, now opt-in with ``PYBIND11_PYTHONLIBS_OVERRWRITE OFF``